An example of why I don't like the PT aggression factor

[MaxPower]

[ QUOTE ]
I think it is odd that you guys are saying AF doesn't mean that much. In NatetheGreat's statistical study, AF was more correlated to winning than any other stat I believe.

[/ QUOTE ]

I don't think it is odd. It is not correlated to winning, it is correlated to win rate. That's not the same thing.

[MaxPower]

Actually if you are talking about the regression that Nate did in this post he did not include AF aggression in it. Is there another one I'm not aware of?

I do believe that AF aggression should be signficantly correlated with BB/100 but that doesn't negate my criticisms of it. I just means that people who bet and raise more after the flop will get better value on their good hands than people who don't. So even if they play bad, they will lose less (unless they are a complete maniac).

[phish]

[ QUOTE ]
[ QUOTE ]
great posts guys. im done with the AF stat.

[/ QUOTE ]

This is the wrong way of reading these posts.

Facts are:

1) As other poster said, with high VPIP they can have low AF and still bet a lot, its just that they call even more.

2) Of course stats (or any read we got on a player) sometimes leads us in the wrong direction. A read doesnt have to be right all of the time, it just has to be right most of the time.

Say I got a read on a player as being extremly passive and he raises me on every street with nothing. Well, the only thing I have to do is write a longer note on him: "Extremly passive but a bit tricky, sometimes doing idiotic bluffs over multiple streets."

We had a good read on him, now we have a better.

[/ QUOTE ]

But how would you have gotten that read unless you were willing to pay him off. If you were willing to do that, it means that you pretty much ignored the AF number, which would make it pretty irrelevant, no?

[bobbyi]

[ QUOTE ]
Two players with the same VP$IP and AF aggresion can have very different styles.

[/ QUOTE ]
No kidding. That's why there are stats other than VPIP and AF. Any two numbers isolated from the rest of someone's game is not going to be enough to describe someone's game, otherwise we would only have two stats available and be done with it. Think about how large the space of playing styles is at holdem. Why would you expect it to be only two dimensional? Most stats (including AF) in isolation are not that meaningful.

It's like if you only looked at one of the cards on the flop when deciding what to do. It wouldn't work very well. But the right conclusion to draw wouldn't be that you should stop looking at the flop at all.

[MaxPower]

[ QUOTE ]
[ QUOTE ]
Two players with the same VP$IP and AF aggresion can have very different styles.

[/ QUOTE ]
No kidding. That's why there are stats other than VPIP and AF. Any two numbers isolated from the rest of someone's game is not going to be enough to describe someone's game, otherwise we would only have two stats available and be done with it. Think about how large the space of playing styles is at holdem. Why would you expect it to be only two dimensional? Most stats (including AF) in isolation are not that meaningful. It's like if you only looked at one of the cards on the flop when deciding what to do. It wouldn't work very well. But the right conclusion to draw wouldn't be that you should stop looking at the flop at all.

[/ QUOTE ]

I don't expect to be able to use 2 numbers to describe someone's game.

I agree that you don't use any one statistic in isolation.

Most of the stats in PT have very simple interpretations. The are just simple percentages. Someone with a high VP$IP plays a lot lot hands. Someone with a low VP$IP plays very tight. I can say with certainty that a player with a VP$IP of 20 players tighter than a player with a VP$IP of 30 (assuming a decent sample size). That is what I want out of a measurement.

The problem with the aggression factor is that I cannot say for certain that someone with an AF of 1 is more aggressive than someone with an aggresion factor of .5 (even if their other stats are the same). It is an issue of measurement validity.

Many of us find the aggression factor useful even though it is not perfect. But why should we settle for that when there may be a better way? I would rather have a less ambigous measurement even if it means I have to think about more things. Looking at the numerator and denominator of the equation seperately would be preferable to me.

[sammy_g]

Have you posted this at the PokerTracker forum? Because I agree, but it isn't likely Pat will see your post here.

[bobbyi]

[ QUOTE ]
Most of the stats in PT have very simple interpretations. The are just simple percentages. Someone with a high VP$IP plays a lot lot hands. Someone with a low VP$IP plays very tight. I can say with certainty that a player with a VP$IP of 20 players tighter than a player with a VP$IP of 30

[/ QUOTE ]
The reason you have "certainty" with the preflop stats is that they aren't dependent on things that happened earlier in the hand. Postflop stats are dependent on how one plays preflop, so they will never be meaningful by themselves in the same way as vpip whether they are percentages or not. For example, went to showdown is a simple %. However, a guy with a vpip of 95 and a wsd of 30 is going to showdown much weaker hands than a guy with a vpip of 12 and a wsd of 30, even though it is a simple percentage. So that alone is not the entire difference between AF and VPIP.

The reason AF is tough to interpret is that there are usually three possible actions postflop: bet/ raise, call or fold. But AF is only the ratio between bet/ raise and call. So without knowing how often the guy folds, you don't know the real percentages. I see your point there. But using other stats, you can get a good picture of how often he folds postflop, so combining that with his AF will give you a good picture of his bet/raise and call percentages. You are right that it might be simpler if (bet %/ call %/ fold %) were commonly used instead of a combination of AF and showdown stats, but they still give the same information (taken together), even if in a slightly convoluted way.

[DcifrThs]

[ QUOTE ]
Actually if you are talking about the regression that Nate did in this post he did not include AF aggression in it. Is there another one I'm not aware of?

I do believe that AF aggression should be signficantly correlated with BB/100 but that doesn't negate my criticisms of it. I just means that people who bet and raise more after the flop will get better value on their good hands than people who don't. So even if they play bad, they will lose less (unless they are a complete maniac).

[/ QUOTE ]

yup.my gut as a statistician tells me AF would be highly correlated w/ win rate.

look at it this way. about 46% of players are pokertracker wise losers.the amount they lose varies from small to huuuuuuuuuge. most winners are fairly marginal. the graph of win rate to AF would be some curvy variation of a 45degree line through the origin shifted to the right.

it may be skewed to acct for the mid range AFs that lose a lot b/c they play too many hands to get the AF up past a certain point. nobody who plays 50% of their hands can have an AF of 2.8 (which is about mine). similarly, nobody who plays 18% of their hands will have an AF of under 1 (because their hands are so good on average compared to their opponents that when they do play they are betting and raising).

since those w/ a 50% vpip are both losers and have "low" AFs and those w/ an 18% vpip tend to be winners with "high" AFs then it follows that AF and win rate would be fairly highly correlated.

the jumble, as it were, would likely occur between 1-2.

i never could get the excel notes file i have to work right (probably b/c i only spent a little while with it), but if somebody would want to look into it it would probably be semi useful as evidence of this conjecture.

-Barron

[kiddo]

[ QUOTE ]
But how would you have gotten that read unless you were willing to pay him off. If you were willing to do that, it means that you pretty much ignored the AF number, which would make it pretty irrelevant, no?

[/ QUOTE ]

If he goes to showdown I will know. And pokertracker will change stats if he does it often enough. But if noone ever calls his raises and he bluffs, u are right, noone will know. But this will only happen if he raises very little and with good hands so...

If Pokertracker says someone got AF 0.5 at flop it means that he calls twice as much as he bets, not using this information because sometimes he will do a stupid bluff makes no sense.